The Eight Words That Decide a Sri Lanka Hotel Booking

Every travel guide tells you to read the reviews. None of them tell you what to look for once you get there. So we took 285,529 written guest reviews of Sri Lankan hotels, guesthouses and homestays and asked a different question than the usual one: not how do people rate this place, but which words in a review only ever appear when something went right — and which only ever appear when something went wrong.

The answer is sharper than we expected. A guest who writes “we will come back” gives a 9 or 10 in 96.6% of reviews. A guest who writes “dirty” gives a 9 or 10 in 5.8%. Across the whole corpus the rave rate is 69.5%. Between those two phrases sits an entire booking decision — and about eight words that act as a free lie detector, if you know which they are.

The one-sentence answer

Read reviews for absence, not praise. A happy guest in Sri Lanka writes “perfect”, “spotless”, “thanks” and “we will come back” — those words are worth almost nothing as evidence because 69.5% of all reviews are already 9s and 10s. What carries information is the reaction that a specific thing went wrong: dirty, broken, mould, overpriced, no hot water, noise. Those words appear in reviews that average 3–5 out of 10 — and their mere presence on a listing is the most reliable signal on the page.

1. The eight words that are gifts

These are phrases that almost never appear in a mediocre review. If a guest wrote one of them, the stay went well enough that the guest reached for language they would not otherwise use. The percentage is the share of reviews containing the phrase that scored 9 or 10.

How to read it. “No hesitation” is the strongest gift in the corpus: it appears in 1,168 reviews and 96.8% of them are 9s or 10s. It is not a description of the hotel at all — it is a description of the guest’s own certainty. The same is true of “will come back” (96.6%) and “unforgettable” (95.6%). These words are not adjectives the guest reached for because the towels were nice; they are conclusions the guest drew.

Below them sit the operational gifts. “Spotless” and “immaculate” appear in 4,843 reviews with a 92.8% rave rate — and they matter because plain “clean” does not. In the same corpus, the word “clean” appears in reviews with a rave rate of just ~78%: guests write “clean” as a neutral box-tick, but they only reach for spotless when they mean it. “Went out of the way” (4,960 reviews, 92.5%) is the same story about people rather than rooms.

2. The words that are traps

These are phrases that almost never appear in a happy review. Their average score is brutal, and the important thing is not the average — it is that the word is there at all.

The veto index, the last column, is the number we think is actually useful. It compares how often the phrase appears in a review scoring 4 or worse against how often it appears in a review scoring 9 or 10. An index of 13.2 for “dirty” means a disappointed guest is thirteen times more likely to have used it than a delighted one. An index below 1 means the word appears in more happy reviews than unhappy ones — which is why it tells you almost nothing.

The practical version: “dirty”, “broken”, “mould”, “no hot water” and “overpriced” are vetoes. One appearance of any of them on a listing page is worth more than fifty five-star one-liners — because it is a specific, verifiable claim that the property failed at something concrete, and because the corpus shows that guests who write those words are almost never writing them about a stay they enjoyed.

3. The most expensive sentence in a review

The words are one thing. This is the pattern that mattered most:

The three words “for the price” appear in 1,485 reviews. Those reviews average 7.08/10 — 1.6 points below the corpus average of 8.64. “Value for money” is worse than it sounds: it appears in reviews averaging 8.62/10, essentially the corpus average — and that is the trap. The phrase is not a complaint. It is the guest starting a calculation.

When someone writes “apart from the bathroom, everything was great”, the resulting score is not an average of great and bad. Across the 204,075 reviews we screened for hedge words, the presence of a single however / although / but / only complaint / apart from drags the review down hard. The reviews that use them and still land at 9 or 10 are a small minority — a guest who volunteers a caveat has usually already decided the caveat matters more than the praise.

4. The test: does the word still work when the review is already happy?

A sceptic could say all of this is circular: of course “dirty” appears in bad reviews, because it is a complaint. So we ran the honest version of the test. We took only the reviews that were already positive — a score of 8 or better — and removed every review containing a hedge word, leaving 205,317 unambiguously happy reviews with a rave rate of 83.3%. Then we asked whether each phrase still moves the outcome inside that happy, caveat-free group.

Three things fall out of that table, and two of them are uncomfortable.

(1) The host is the single most valuable thing in a Sri Lankan review. Even among guests who already gave a positive score, the ones who wrote about the host rave at 92.6% against 82.7% for those who never mentioned a person — a 9.9-point gap. Nothing about the building comes close. If you are choosing between two similarly rated properties, read for the host.

(2) Wi-Fi and breakfast are not what people think they are. Among already-happy guests, mentioning Wi-Fi moves the rave rate down slightly (80.2% vs 83.3%) — the word appears in a review only when someone bothered to evaluate it, and evaluation is a mild negative. Breakfast stays modestly positive (87.6% vs 82.6%). Neither is the differentiator the marketing suggests. The bathroom is.

(3) “Dirty” cannot be rescued. Among guests who gave 8 or better and wrote no caveat at all, the 120 reviews that still contain “dirty” or “unclean” rave at only 45.0% — a 38.3-point collapse inside a group that was otherwise 83.3% delighted. A guest who says the room was dirty and scores the stay 8 anyway is not a satisfied guest. They are being polite.

5. What this looks like at the property-type level

The same eight words behave differently depending on whether you are looking at a hotel or a homestay — because the guests are using different words to describe the same good experience.

Homestays are the happiest category on the island (81.3% rave rate against 67.0% for hotels) and they also talk about their hosts more: 17.9% of homestay reviews mention a host, against 5.6% of hotel reviews. Guests in a homestay are reviewing a person. Guests in a hotel are reviewing a service. Both can be excellent; only one of them is likely to make you write “unforgettable”.

6. How to use this in ten minutes

Practical, in the order you would actually do it:

1. Sort by worst, not by best. Every booking site lets you. Read the one-star reviews first and look only for the veto words: dirty, mould, broken, no hot water, overpriced. If none appear across twenty negative reviews, the property has a standard.
2. Then read the middle. The 6s and 7s are where guests explain. Look for “for the price” and for “however” — both mean the guest kept a ledger, and the ledger usually ends in a deduction.
3. Count hosts. Scroll the recent ten reviews and count how many mention a person by role. If four or more do, the human side of the stay is the product, and the ratings will hold up under pressure. Our breakfast study found the same shape: homestays out-score hotels by 0.68.
4. Ignore “clean” and trust “spotless”. A review that says “clean” is ticking a box. A review that says “spotless” or “immaculate” means someone was surprised.
5. Treat “will be back” as a real signal and “would recommend” as a weak one. 2,235 reviews contain an explicit intention to return and 96.6% of them are 9s and 10s. Recommendation is broader and therefore cheaper.

For the wider set of patterns behind these numbers, the longer-running studies in this series cover how rarely Sri Lankan hotels answer reviews (only 1 in 10), whether the Wi-Fi actually works, what a single hotel towel predicts, and how the two biggest booking sites disagree about the same hotels.

7. What this study cannot tell you

It is correlation, not causation, and the direction is partly circular. “Dirty” predicts a low score partly because the guest is telling you, in a word, that the room was dirty. The value is not that the word causes anything — it is that the word is specific. A guest who writes “dirty” has made a factual claim you can check. A guest who writes “disappointing” has told you how they felt, which is not checkable and not transferable.

The corpus is saturated at the top. 69.5% of these reviews score 9 or 10 and only 7.0% score 4 or worse. That is why we report rave rates and veto indices rather than average scores: on a scale where the average is already 8.64/10, a 0.2-point difference is statistically real and practically meaningless. Rave rate makes the differences visible.

Language and culture move the words. The corpus spans reviewers from many countries and several languages, and the same stay generates different vocabulary depending on who is writing. That is one reason we kept to phrase-level patterns rather than single words wherever possible: “will come back” travels better across languages than “lovely” does.

Sample sizes vary. Some phrases here rest on 150 reviews and some on 20,000. Every table shows the count for exactly this reason. Where a row rests on fewer than 400 reviews we say so in the count rather than smoothing it away.

Frequently asked questions

The bottom line

Across 285,529 written reviews of Sri Lankan stays, the words that predict a good stay are not the adjectives you would guess. They are the phrases guests only use when they have made up their mind: no hesitation, we will come back, spotless, went out of the way. The words that predict a bad one are specific, not emotional: dirty, mould, broken, overpriced, no hot water. And the single most valuable thing a guest can mention is not the pool, the breakfast or the Wi-Fi — it is a person, which lifts even an already-happy review by 9.9 points.

Read for specifics. Ignore the praise, hunt the vetoes, and count the hosts.

Method: all 285,529 verified guest reviews in the byc.lk corpus that contain at least 40 characters of written text, drawn from 363,148 reviews in total. Score scale 1–10. Rave = score of 9 or 10. Veto index = (share of 4-or-worse reviews containing the phrase) ÷ (share of 9-or-10 reviews containing the phrase); values below 1 mean the phrase is more common in happy reviews. Phrases are matched case-insensitively with regular expressions on the review body; single-word patterns match any occurrence. The held-out test in section 4 uses only reviews scoring 8 or better with no hedging marker (however, although, but, though, apart from, only complaint, one thing, minor, everything was great/perfect). Property types as recorded by the source. Last updated: 23 October 2026.